70 lines
1.9 KiB
Markdown
70 lines
1.9 KiB
Markdown
---
|
|
language:
|
|
- en
|
|
- bn
|
|
license: mit
|
|
library_name: transformers
|
|
tags:
|
|
- medical
|
|
- bangladesh
|
|
- qwen
|
|
- gguf
|
|
- healthcare
|
|
- offline
|
|
- mobile
|
|
pipeline_tag: text-generation
|
|
model-index:
|
|
- name: ShasthoGPT-BD-3B
|
|
results: []
|
|
datasets:
|
|
- Mahadih534/all-Bangladeshi-medicines
|
|
- GBaker/MedQA-USMLE-4-options
|
|
base_model:
|
|
- Qwen/Qwen2.5-3B-Instruct
|
|
---
|
|
|
|
# 🏥 Qwen-MediCare-BD
|
|
|
|
**Bangladesh's First Offline Medical AI Assistant**
|
|
|
|
## Model Description
|
|
|
|
Qwen-MediCare-BD-3B is a fine-tuned medical language model based on Qwen2.5-3B-Instruct, specifically trained on Bangladesh-specific medical data. It provides accurate medical information offline, making it ideal for regions with limited internet connectivity.
|
|
|
|
### Key Features
|
|
- 🇧🇩 **Bangladesh-specific**: Includes local diseases, drugs, and medical context
|
|
- 📱 **Mobile-ready**: Quantized to Q4_K_M (1.8GB)
|
|
- 🔒 **100% Offline**: No internet required for inference
|
|
- 🩺 **Medically validated**: Trained on 30,523 medical Q&A pairs
|
|
- 💚 **Multilingual**: Supports English and Bangla queries
|
|
|
|
## Model Variants
|
|
|
|
| Variant | Size | Format | Use Case |
|
|
|---------|------|--------|----------|
|
|
| Full Model | 6.2 GB | Safetensors | Training/Research |
|
|
| Q4_K_M | 1.8 GB | GGUF | Mobile/Edge devices |
|
|
| LoRA Adapters | 114 MB | Safetensors | Fine-tuning |
|
|
|
|
## Quick Start
|
|
|
|
### Using Transformers
|
|
```python
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer
|
|
|
|
model = AutoModelForCausalLM.from_pretrained(
|
|
"CBrootA/Qwen-MediCare-BD",
|
|
device_map="auto",
|
|
load_in_4bit=True
|
|
)
|
|
tokenizer = AutoTokenizer.from_pretrained("CBrootA/Qwen-MediCare-BD")
|
|
|
|
messages = [
|
|
{"role": "system", "content": "You are a medical assistant for Bangladesh."},
|
|
{"role": "user", "content": "What are dengue symptoms?"}
|
|
]
|
|
|
|
inputs = tokenizer.apply_chat_template(messages, return_tensors="pt").to("cuda")
|
|
outputs = model.generate(inputs, max_new_tokens=256)
|
|
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
|